CI Artifact And Test Temp Disk Full Policy

When CI fails with No space left on device after tests pass or near artifact upload, the root cause is often peak temporary storage: test outputs, binlogs, fixture fanout, Docker image archives, Playwright/browser artifacts, or a scratch directory that is never pruned between shards.

Runbook: Budget The Upload And Scratch Peaks

1. Measure disk before tests, before artifact upload, and after upload. Many failures happen after the test result exists.
2. Track each mount separately: root, workspace, /tmp, Docker root, tool cache, and runner temp.
3. Classify outputs as reusable cache, generated test scratch, required failure evidence, or upload-only artifacts.
4. Prune generated scratch only after the shard has uploaded the evidence needed for debugging.
5. Add a hard free-byte reserve before the next shard, Docker image archive, or artifact upload starts.
6. When failing early, report the largest bucket and the first safe cleanup action.

I would treat this as a CI peak-storage budget problem, not just generic runner cleanup.

The useful evidence is:
- free bytes before tests
- free bytes before artifact upload
- top workspace/temp/report/artifact directories
- whether Docker/kind/image archives create a second full copy
- which outputs are required for failure diagnosis versus generated scratch

Acceptance rule: every shard should leave enough reserve for the next shard or upload step, and cleanup should only remove generated outputs after required logs/artifacts are uploaded.

Do Not Delete First

• Failure logs, binlogs, screenshots, traces, coverage files, or minidumps that are the only evidence for the failed run.
• Tool caches on shared or self-hosted runners before confirming they are rebuildable and not shared with active jobs.
• Docker volumes or named caches unless the job created them and the cleanup boundary is explicit.
• Source workspace state before artifact upload has captured the needed debug output.